December 4, 2023. Another day in the life, another set of challenges and opportunities. The past year has been a whirlwind with AI/LLM infrastructure continuing its meteoric rise, platform engineering becoming mainstream, and FinOps driving cloud cost pressure like never before. I find myself reflecting on a particularly busy period and how it shaped my work.

First up, debugging a tricky issue with Gemini AI. We’ve been working closely with them, and their LLM (Language Learning Model) was showing some unexpected behavior in certain high-traffic scenarios. The problem boiled down to an edge case where the model was getting overloaded due to synchronous processing bottlenecks. After hours of tracing logs and profiling, we identified the culprit—a misconfigured rate limiter that was failing under load. It’s a humbling reminder that even with all the advanced tools at our disposal, fundamental engineering principles still matter.

On the platform side, we’ve been heavily invested in WebAssembly (Wasm) for server-side use cases. The idea of running JavaScript code on the backend has been fascinating and challenging. We recently shipped a microservice written entirely in Wasm, and it’s performing well. However, there were some growing pains, particularly around debugging and optimization. It feels like we’re still figuring out the best practices for building production-grade applications with this technology. The developer experience is improving, but there’s no denying that working within the constraints of a sandboxed environment can be frustrating.

Platform engineering has definitely become mainstream. I’ve been attending more meetings than coding lately—discussions about FinOps and cloud cost optimization are now routine. We’re moving towards more transparent cost reporting, which means I need to keep an eye on our spend. Last week, we hit some budgetary limits due to a misconfigured auto-scaling group in AWS. It’s a good reminder that even the most straightforward infrastructure changes can have significant financial impacts if not properly managed.

Speaking of FinOps, there’s been a lot of buzz around DORA metrics recently. Our team has adopted them, and it’s been eye-opening. We’re tracking deployment frequency, lead time for changes, change failure rate, and mean time to recovery (MTTR). These metrics are giving us insights into where we can improve our processes. For example, one area that stands out is the time it takes to bring down a service when something goes wrong. I’ve been arguing internally for more robust automation to reduce MTTR, and while some progress has been made, there’s still a long way to go.

One of the projects I’m excited about is integrating with Figma. We’re trying to streamline our design process by automating some of the hand-offs between designers and developers. It’s early days, but so far it looks promising. The fact that Adobe abandoned its proposed merger with Figma adds a layer of complexity—now we have two major players in the space, each pushing their own tools and workflows.

And let’s not forget about the challenges posed by platform engineering tracks normalizing. At our company, there’s been an increase in demand for specialized roles like DevOps engineers and platform architects. This has led to some healthy debates about whether these are new roles or just evolutions of existing ones. Personally, I feel that while the titles might change, the core responsibilities remain the same—ensuring robust, scalable, and maintainable infrastructure.

Reflecting on today, it’s clear that we’re in an exciting but challenging time. AI/LLMs continue to push the boundaries of what’s possible, while FinOps and cloud cost pressure keep us focused on efficiency. It’s a delicate balance, but one I’m happy to be part of. Here’s to another year of learning, growing, and pushing the limits of what we can achieve.

---

This isn’t just about the tech; it’s about the people behind the scenes making it work day in and day out. Happy holidays!